The study of prompt and non-prompt production of charm hadrons is crucial to test the limits of perturbative QCD and to understand the beauty hadron production in collider experiments. In this contribution, we propose a machine learning (ML)-based method to separate the prompt and non-prompt production of open ( \(\mathrm D^{0}\) ) and hidden charm ( \(\textrm{J}/\psi \) ) hadrons. We employ XGBoost, LightGBM, Cat Boost, etc., ML models which take track-level information for training and prediction. For the training, we reconstruct \(\mathrm{D^{0}\rightarrow \pi ^{+}K^{-}}\) and \(\mathrm{J/\psi }\rightarrow \mu ^{+}\mu ^{-}\) decay channels. Further, invariant mass, pseudo-proper decay length, distance of closest approach, proper time, pseudorapidity, and transverse momentum of the decay candidates are used for the training and predictions. We obtain about 99% accuracy in the prediction from the ML models.